Romans1by1: a case study on graph-based modelling of prosopographical datasets

Romans 1by1 is a digital-born prosopography focusing on people attested in classical era inscriptions; it presently hosts approximately 15.000 open access persons files. In the scope of this paper, we use Romans1by1 (http://romans1by1.com/) as an example to explore the perspectives of ingesting the information from a prosopographical relational database into a graph database.

Prosopography is a branch of historical research that emerged during the 19th century and soon became of major interest for the researchers of Antiquity. While it was initially oriented towards well known elite personalities of the past, on whom information is rich, it has nowadays turned more and more towards the regular inhabitants of the ancient world, thus going into the depths of the socio-economic fabric of societies, revealed through micro-networks and the connections between them. 

In the digital era, building prosopographies for the ancient world is very much tributary to the model suggested by John Bradley and his research group at King’s College London, which marked the shift from narrative descriptions to information on persons as structured data (https://www.kcl.ac.uk/factoid-prosopography). Working mainly on the Prosopography of Anglo-Saxon England (PASE), Bradley theorized the concept of prosopographical “factoids” – small pieces of data about persons (name variants, occupations, places of residence, events in which they played a role, personal and professional relationships, possessions, etc.) provided by the sources, which are not exactly facts as we cannot verify their veracity. Factoid prosopographers took these pieces of information and used to put them into a relational database that reflected their connections and relationships. One of the main advantages of this model lies in the fact that it offers the opportunity to stay true to the source – a necessity when we are sometimes dealing with inconsistent or even antagonistic sources. 

Ever since, digital prosopographies typically followed Bradley’s methodological framework, and most of them are built as relational databases using SQL or similar query languages. Romans1by1 is no exception in this regard and has been designed as a relational database, meant to include all known individuals dating from the Greek and Roman periods (exhaustively up to the end of the 3rd c. AD), from the European provinces of the Empire. The database is meant to be user-friendly and adaptable to various research questions, most notably through search filters in each main category which allow the user to query a specific set of information from the database. By this means, the user can rapidly find out, for example, who were all members of the collegium tignuariorum from Arelate, or how many musicians are attested in the Western provinces. Moreover, gathered information can be downloaded in a spreadsheet file and processed by other software.

Though recent research in prosopography conceptually continues to use the factoid model, there is a trend toward exploring new approaches and new software. One example in this context is SPEAR, a project which translates factoid data into XML schemata. In this paper, we present and discuss another promising research avenue, that is the use of graph-based models and software for prosopographical data sets.

Network models offer a powerful way to understand and analyze complex data by relating the data items to nodes and edges, the edges representing the relationships between the nodes and their properties. Some scientific fields have been using network models for a long time, most notably systems biology where the graph approach is being developed for more than twenty-five years. A number of software are readily available and widely used for graph analysis and visualization, such as Gephi or Cytoscape, where data coming from any data source can be imported via graph-formatted files. However, as the spreadsheets have to be structured in such a manner as to answer certain specific research questions, the possibilities for ‘open questions’ are reduced, and the effort of creating compatible datasheets is often not proportionate with the relevance of the obtained visualizations, especially when dealing with very large datasets. 

Therefore, there is today a movement in the DH research community toward using dedicated graph databases where information is from the outset inherently stored in the form of a graph. These databases also typically come with built-in tools, such as the Neo4j database shipped with the Neo4j Browser, allowing the user to inspect their dataset in an interactive way, via database queries and simple visualisations of the results. Crucially important when dealing with historical information datasets, as it is the case with Romans1by1, is the possibility to work with a graph database management system and not a mere visualization tool. This assures the real-time synchronization between dataset and visualization and allows the possibility to keep research dynamic. Graph visualization platforms like Intergraph, connected to such databases, offer even more convenient ways to explore the data by a lightweight search and discovery workflow and interest-driven navigation, especially for users having no or very limited experience with database manipulation and data analysis. While the whole dataset is ingested, the user can focus on a subset of relevant information through a scene of subgraphs which can be dynamically modified and interconnected. 

This paper takes Romans1by1 as an example to present the general procedure of migrating from a relational database to a graph database, and to discuss the value of this approach through a series of insights which would have been difficult to obtain by conventional exploration methods in relational databases. Relating to an example suggested above, after finding out who all the attested musicians from the Roman provinces were, a graph perspective could also visualize each musician’s personal networks (familial, professional, social) and the indirect connections (the relations of the direct networks’ members), as well as the connection between types of sources, the time span of attestations and their geographic and diachronic distributions. 

At the present time, our work is still in progress, and not all potentially relevant individual case studies have yet been explored.  Nonetheless, the main outcomes can be divided in three general categories: source-connected, network-connected and diachronic distribution graphs. The relation of a given person to the sources and how he/she is represented in various texts is highly important. Going a step further than the factoid model, we textualize individuals and individualize texts, meaning that the inscription itself - and the ‘factoids’ it narrates - become part of the individual’s ego-history (the methodology comes from the Lived Ancient Religion group: https://www.uni-erfurt.de/en/max-weber-centre/projects/cooperation-projects/lived-ancient-religion/). Thus, the link between individual and source becomes more important than initially imagined. The network analyses refer to individual networks, but it also allows researchers to compare and superpose two (or more) persons’ networks, as well as to visualize and investigate ‘immaterial’ networks, connections between institutions, economic networks, etc. The diachronic distribution graphs show quantitative and qualitative changes that occur over time at the scale of epigraphic habits, network patterns and individual movement. Correlating all these types of information and visualizing them comes with great analytic gains, as it reveals a more complex picture on the people we study and the fabric of their society.

The proposed paper will showcase the methodological steps and complexities of transposing a relational prosopography into the graph database model, as well as the scientific benefits of such an endeavour. While Romans1by1 has its undeniable specificities, imprinted by the nature of its sources, our work can also serve as a model-demonstration for other digital prosopographies.